Adapting a developed market credit risk model for the understanding and estimation of consumer credit losses in South Africa

Abstract

The aim of this research was to build a developed market credit risk model and adapt it for the understanding and estimation of consumer credit losses in the emerging market of South Africa. The developed market of the United States of America was chosen since it has good quality data (the research is quantitative in nature). Credit, a contractual agreement in which a person or institution that is a party to the agreement, borrows something of value today with an undertaking to repay (often with interest) the other party at a later date(s), has existed and evolved for thousands of years. With almost every financial transaction comes credit risk. Credit risk is measured using credit risk models and reviewed literature indicates that models that have been developed have poor estimation accuracy levels. Model errors are important as they directly affect profitability, solvency, shareholder value, macro-economy, and society. From publicly available data at big databases such as the FRED (Reserve Bank Economic Data (US)), the SARB (South African Reserve Bank) and the World Bank, independent variables (with data of monthly and quarterly frequencies and spreading from 2008-2021 and 1987-2021 for SA and the US respectively) were selected. Multivariable regression analysis, amongst other analyses, was done to establish relationships between individual or sets of the selected independent variables and credit losses. Explanatory variables that capture Sentiment (defined in this research as reactions or behaviour of credit market participants in far-from-equilibrium situations - for example recessions, financial booms and busts) were coupled with economic variables and obligor characteristics. The establishment of a model building analyses blocks framework and a relatively accurate consumer credit risk model for the emerging market of South Africa (with an R-squared value of 85% and back test estimation accuracy of 88% on average) were amongst the key results of the analyses. The universal applicability, subject to the availability of quality data, of the credit risk modelling methods used in this research could motivate policymakers in South Africa and other emerging markets to adopt the data collection, storage, and organisation formats used in the FRED database.